| Benchmark | DeepSeek-R1 | Gemini 2.5 Flash |
|---|---|---|
| AA Agentic Index | 3.1 | 18.8 |
| AA Intelligence | 18.6 | 17.7 |
| AA-LCR | 56 | 65.7 |
| AA-Omniscience | -31.3 | -29.8 |
| aider_polyglot | 71.6 | 61.9 |
| AIME 2024 | 91.4 | 82.3 |
| AIME 2025 | 87.5 | 72 |
| AIR-Bench 2024 | 52.9 | 68.7 |
| anthropic_red_team | 99.1 | 98.8 |
| arc_agi_1 | 21.2 | 33.3 |
| ARC-AGI-2 | 1.3 | 2.5 |
| Artificial Analysis Coding Index | 24.6 | 22.2 |
| bbq | 96.6 | 97.7 |
| critpt | 0.6 | 1.4 |
| gdpval | 1.5 | 11.9 |
| GPQA Diamond | 81 | 82.8 |
| harmbench | 54.6 | 62.6 |
| HLE | 17.7 | 12.1 |
| hmmt_feb_2025 | 76.7 | 64.2 |
| IFBench | 39 | 50.3 |
| LiveCodeBench | 84.4 | 76.2 |
| livecodebench_easy | 99.2 | 99.1 |
| livecodebench_hard | 63.6 | 48.3 |
| livecodebench_medium | 90.9 | 82.5 |
| OmniScience Accuracy | 30.7 | 26.1 |
| OmniScience Non-Hallucination | 10.5 | 24.7 |
| scicode | 35.7 | 35.9 |
| simple_safety_tests | 98.3 | 98 |
| simplebench | 40.8 | 41.2 |
| simpleqa | 92.3 | 26.9 |
| SWE-bench Verified | 57.6 | 60.4 |
| Terminal-Bench Hard | 6.1 | 13.6 |
| vectara_answer_rate | 97 | 99 |
| vectara_avg_summary_length | 93.5 | 101.5 |
| vectara_factual_consistency | 88.7 | 92.2 |
| vectara_hallucination_rate ↓ | 11.3 | 7.8 |
| xstest | 98.8 | 98.8 |
| τ²-Bench Telecom (AA run) | 11.4 | 31.6 |
Best tracked score per model per benchmark (default configuration; source-attributed). ↓ marks lower-is-better metrics. Open either model for its full surface, provenance and pricing.